© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    APIAssistant AxisNEWCircuit TracerNEWSteerSAE EvalsExports Community BlogPrivacy & TermsContact
    1. Home
    2. Google DeepMind · Exploring Gemma 2 with Gemma Scope
    3. Gemma-2-9B-IT
    4. Residual Stream - 131k
    5. 31-GEMMASCOPE-RES-131K
    6. 64531
    Prev
    Next
    INDEX
    Explanations

    instances of contrastive conjunctions and signaling phrases

    oai_token-act-pair · gpt-4o-miniTriggered by @bot

    introducing contrast "but"

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GEMMA-2-9B-IT @ 31-gemmascope-res-131k
    Configuration
    google/gemma-scope-9b-it-res/layer_31/width_131k/average_l0_109
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    Features
    131,072
    Data Type
    float32
    Hook Name
    blocks.31.hook_resid_post
    Hook Layer
    31
    Architecture
    jumprelu
    Context Size
    1,024
    Dataset
    monology/pile-uncopyrighted
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
     yeah
    -0.66
    Anyways
    -0.65
     Anyways
    -0.63
     anyways
    -0.62
     rightfully
    -0.59
    Honestly
    -0.57
     yells
    -0.57
     tryna
    -0.57
     Appreciate
    -0.57
     kinda
    -0.56
    POSITIVE LOGITS
     ciascuno
    0.58
     although
    0.54
     Daß
    0.50
    こうした
    0.49
     ilustracja
    0.47
     daß
    0.46
     läßt
    0.45
    事实上
    0.45
    although
    0.45
     sekal
    0.45
    Activations Density 0.069%

    No Known Activations